JP2003131696A - Voice registration authentication system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To increase the authentication speed and authentication rate of a registration type voice authentication system by classifying a voice by its feature when the voice is registered and retrieving approximate data from data in the closest classification group according to the feature of the voice when authentication is performed. SOLUTION: The system in which an examinee registers a voice and authenticates a speaker has a means which extracts individual differences of an input voice and classifies the voice by the feature when the voice is registered, a means which registers (stores) data by classifications, a means which extracts the feature of a voice when voice authentication is performed as well as the registration, a means which authenticates which classification frame of a voice database matches the input voice for authentication according to the feature of the voice, a means which retrieves approximate data from an estimated classification frame, a means which finds an authentication result, a means which makes the examinee confirm the authentication result, and a means which performs re-retrieval by judging whether the authentication result is correct or incorrect.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is used in a registration type voice recognition system for the purpose of speaker authentication, when it is necessary to improve the accuracy of an authentication rate or obtain an authentication result as soon as possible. A voice registration authentication system that can be used.

[0002]

2. Description of the Related Art In the conventional registration-type voice recognition method, voice data input at the time of registration is attached with a label such as an ID and stored in a database or the like, and at the time of authentication, similar data is searched for and authenticated. In most cases, the voice was converted to text and registered, and at the time of authentication, a text with similar pronunciation was searched and the result was obtained.

[0003]

In the present invention, at the time of voice registration, voice characteristics are classified and registered (saved) in a database for each group. Next, at the time of authentication, the input voice is classified and similar data is searched for only from the optimum classification frame (group) in the database to obtain the authentication result. By these processes, it becomes possible to proceed with the process without requiring the subject to perform a special operation. In addition, it is possible to increase the authentication rate by performing a search from the optimum classification frame at the time of authentication, and at the same time to increase the speed of obtaining an authentication result by reducing the amount of data to be searched.

The first problem to be solved by the present invention will be described below.

At the time of voice registration, a group of speakers (men and women / adult children, input keyword length, included) is included according to voice characteristics (for example, difference in frequency distribution or pitch, length of included sound or word). Different according to the sound that is played, etc.) Search for approximate data from the optimal group based on the characteristics of the voice input at the time of authentication,
It is to provide a system capable of increasing the certification rate and the certification speed by obtaining the certification result.

The second problem to be solved by the present invention will be described below.

At the time of voice registration, speakers are classified into groups and stored in a database. For example, assume a hospital, a factory, an office, a laboratory, etc., which has introduced a flextime system or a three-shift work system for midnight and nighttime, and which performs a security check by voice when entering and exiting. People who go in and out of these facilities (go to work and leave) are called members.
If the members have a tendency to go to and from the office, the schedule is managed in advance, or the time to leave the office can be predicted from the office time, the grouping based on the time for each member becomes possible at the time of voice registration. Therefore, the voice registration data is classified into groups in the voice database and stored.

Since the attendance time and the leaving time can be classified for each member in this way, the member data is retrieved from the classified groups at the time of authentication.

Therefore, a second problem to be solved by the present invention is to provide a system capable of increasing the authentication rate and the authentication speed.

[0010]

The present invention has solved the above-mentioned problems by having the following constitution.

(1) In the registration-type speaker authentication system, a means for deriving a voice feature (individual difference) based on the frequency domain of voice, the length of a keyword to be input, the included sound element, etc. at the time of voice registration; Means for classifying by characteristics, means for storing data in the classified groups, means for extracting voice characteristics during voice authentication, and means for deciding from which classification frame the registered data should be retrieved by voice characteristics And a means for retrieving an approximate value from the determined classification group, a means for outputting an authentication result, a means for confirming the authentication result with a subject, and a means for re-searching from the correctness of the authentication result. Voice registration authentication system to do.

(2) In the registration-type speaker authentication system, when the registration time and the authentication time can be managed by a schedule, a means for classifying the voice of the input person according to the schedule at the time of voice registration, and storing the data in the classified groups. Means, a means for classifying voices according to the time at the time of voice authentication, a means for deciding from which classification frame the registered data is searched, a means for searching an approximate value from the decided classification group, and an authentication result. A voice registration / authentication system comprising: a means, a means for confirming an authentication result with a subject, and a means for performing a re-search based on whether the authentication result is correct or incorrect.

[0013]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings.

<Description of Block Diagram of FIG. 9> FIG. 9 is a block diagram for explaining the system configuration of the present invention.

Reference numeral 1001 in the figure is a CPU for controlling the present system, and 1002 is an input unit for inputting voice, for example, a microphone is equivalent thereto. From here, the raw voice 1007 is input and taken into the computer, 100
Reference numeral 3 is a storage device. The raw voice 1007 input from the input unit 1002 is converted into an appropriate voice model 1008 by the control of the CPU 1001 and stored in the storage device 1003. The voice database etc. are also stored here. 100
A RAM 4 provides a work area to the CPU 1001 and loads an application. Reference numeral 1005 is an input unit for inputting the subject's will in the process of proceeding interactively. For example, a mouse corresponds to this. 10
An output unit 06 displays the flow of conversational processing. For example, a monitor corresponds to this.

Example 1 <FIG. 1 Regarding Conceptual Diagram of Registration Process of Example 1> FIG. 1 is a conceptual diagram illustrating a process at the time of registration of Example 1 of the present invention.

In the figure, squares represent data, and bold squares represent processing. White arrows (104, etc.) show the flow of data, and black arrows (107, etc.) show the flow of processing. A series of processes is controlled by the CPU 1001 in FIG.
Reference numeral 101 is voice data before processing of the keyword input at the time of registration. Computer 10 through microphone 102
Get stuck in 3. The raw voice is voice modeled by the computer 103 (105). The input voice is feature-extracted (106), classified (108), registered in the voice database 110 (109), and stored in the storage device 1003 of FIG. In this schematic diagram, for example, the female group (111) is classified by frequency analysis.

<FIG. 2 Regarding Conceptual Diagram of Authentication Process of First Embodiment> FIG. 2 is a conceptual diagram illustrating a process at the time of authentication according to the first embodiment of the present invention.

In the figure, squares represent data, and bold squares represent processing. White and dotted arrows (204, etc.) show the flow of data, and black arrows (207, etc.) show the flow of processing. A series of processes is controlled by the CPU 1001 in FIG. Reference numeral 201 is voice data before processing of the keyword input at the time of registration. It is taken in by the computer 203 through the microphone of 202. Raw voice is computer 203
Voice modeled by (205). The input voice is feature-extracted (206), and the voice database (21
The optimum group is estimated from the groups classified into 0) (208), and the authentication result is searched from them (209). In the example of the figure, the input voice is estimated to be female by feature extraction, and approximated data is retrieved from the female group (211) in the voice database (211) and the retrieval result is output (212).

<Description of Flowchart of Embodiment 1 in FIGS. 5 and 6> A flow for carrying out the present invention will be described with reference to FIGS.

<Description of Process Flow in Voice Registration Unit of First Embodiment> FIG. 5 shows a process performed by a computer unit (103 in FIG. 1) which is a control unit of the voice registration process.

Step (501) The subject inputs an arbitrary keyword for registration.

Step (502) The input raw voice is digitized and voice modeled for storage in a voice database.

Step (503) The feature of the voice is extracted. For example, as shown in the figure, the cumulative distribution model of the amplitude level of the voice, the statistical distribution model of the fundamental frequency, the model based on the content ratio of consonants and vowels, the model based on the content ratio of unvoiced and voiced sounds, and the keyword Create a model based on time length, etc.

Step (504) Classify the current voice data into the optimum group in the voice database. For example, based on the model created in the previous step, it is classified whether the keyword is Japanese or another language, and whether the keyword is male voice, female voice, adult or child.

Step (505): Save in the voice database for each classification.

In the above flow chart, the voice feature extraction and classification means in steps (503) and (504) are the characteristic processes of the present registration unit.

<Description of Process Flow in Voice Authentication Unit of First Embodiment> FIG. 6 shows a process performed by the computer unit (203 in FIG. 2) which is the control unit of the voice authentication process.

Step (601) The subject inputs the same keyword as at the time of registration for authentication.

Step (602) The input raw voice is digitized and voice modeled for storage in a voice database.

Step (603) The feature of the voice is extracted.

Step (604) The group in which the data corresponding to the current voice data is considered to be registered is specified.

Step (605) Similar voice data is searched from the specified group.

Step (606) The ID of the most similar data is returned as a return value.

Step (607) Confirm whether the authentication result is correct. If it is correct, the authentication process ends. If it is wrong, the process returns to step (604) to perform the group identification process.

In the above flow chart, the voice feature extraction and classification group identification means in steps (603) and (604) are the characteristic processes of the authentication unit of the present embodiment.

Second Embodiment <FIG. 3 Regarding Conceptual Diagram of Registration Process of Second Embodiment> FIG. 3 is a conceptual diagram illustrating the registration process of the second embodiment of the present invention. Here, an example of a system that manages the departure time of a hospital will be described. At this hospital from 8 am to 4 pm
There are three shifts: "morning-evening work" at night, "day-night work" between 4:00 pm and midnight, and "midnight work, shift duty" between midnight and 8:00 am ing. The work schedule of each worker is determined monthly. The voice of the worker is registered in the computer in advance and is managed together with the monthly schedule. This process will be described with reference to FIGS.

A series of processing is performed by the CP in the computer 303.
Controlled by the U1001, the RAM 1004 is used as a work area, the work schedule data (312) is loaded into the storage device 1003, and the voice model is stored. Further, the display or confirmation work for notifying the progress of the process is performed through an input unit 1005 such as a mouse and a touch sensor and an output unit 1006 such as a monitor in FIG.

In the figure, white arrows (304 etc.) show the flow of data, and black thick arrows (307 etc.) show the flow of processing. Each worker inputs the keyword voice 301 through the voice input device 302-microphone 302. The voice is taken into the computer 303 and converted into the voice model 305 under the control of the CPU. The voice model is a monthly work schedule 3 that has been entered in advance as data.
Feature extraction (306) by 12 (determines which working time zone it belongs to, and groups for each time period (313)
(308). This voice registration process is finally performed in the voice database (310) in the storage device 1003 of FIG.
It is saved in the state (311) classified as.

<FIG. 4 Regarding Conceptual Diagram of Authentication Process of Second Embodiment> FIG. 4 is a conceptual diagram illustrating the voice authentication process of the second embodiment of the present invention.

A series of processing is performed by the CP in the computer 403.
It is controlled by the U1001 and uses the RAM 1004 as a work area to load the work schedule data (412) in the storage device 1003 and store the voice model. Further, the display or confirmation work for notifying the progress of the process is performed through an input unit 1005 such as a mouse and a touch sensor and an output unit 1006 such as a monitor in FIG.

In the figure, a white arrow (404 or the like) indicates a data flow, and a black thick line arrow (407 or the like) indicates a processing flow. First, it is assumed that the present time is a midnight / shift duty zone for workers who are dispatched around midnight. B
A case where a nurse comes to the hospital will be described as an example.

The nurse B inputs the same keyword as that input at the time of registration to the computer 403 through the input device-microphone of 402. The voice 401 is the computer 40
It is converted into the voice model 405 by the process of 3. It is assumed that the current time is around 4:00 pm, that is, the dispatch time of the midnight / shift duty zone (412, 406, 408). If an arbitrary person inputs a voice at work here, the authentication process searches the voice database (410) for similar data from the data registered in the late night / shift duty zone group (409). ). In other words, if nurse B goes to work as scheduled, it will be in the midnight / shift duty group (41
The registration data (411) of nurse B in 3) will be found.

<Explanation of Flowchart of Embodiment 2 in FIGS. 7 and 8> With reference to FIGS. 7 and 8, a flow for implementing the present invention will be described.

<FIG. 7: Description of Process Flow in Voice Registration Unit of Embodiment 2> FIG. 7 shows the process performed by the computer unit (303 in FIG. 3) which is the control unit of the voice registration process.

Step (701) The work hour schedule of the worker for each month is input.

Step (702) For registration, the employee speaks and inputs an arbitrary keyword.

Step (703) The input raw voice is digitized and voice modeled for storage in a voice database.

Step (704) The classification group of the voice input person is specified by referring to the work hours schedule.

Step (705) classify the voice models into the specified groups.

Step (706) Save to voice database.

In the above flow chart, steps (701) and (704) are the characteristic processing of this embodiment-voice registration section.

<Description of Process Flow in Voice Authentication Unit of Second Embodiment> FIG. 8 shows a process performed by the computer unit (403 in FIG. 4) which is the control unit of the voice authentication process.

Step (801) The worker who goes to work at the hospital inputs the same keyword as at the time of registration by voice input for authentication.

Step (802) The input raw voice is digitized and voice modeled for storage in a voice database.

Step (803) Get the voice input time of the worker who went to work.

Step (804) From the time obtained in the previous step, it is estimated which working time zone the dispatch time corresponds to.

Step (805) Data similar to the voice model is retrieved from the group of working hours estimated in the previous step.

Step (806) Obtain the most similar search result.

Step (807) The subject is checked for correctness of the authentication result. If it is correct, the authentication process ends.
If wrong, go to the next step (808).

Step (808) It is confirmed whether or not all items have been searched in the classification group of one working hours. If all the data in the group have been retrieved, the process proceeds to step (809). If data remains in the group, the process returns to step (805) to continue the search process for the next similar candidate.

Step (809) When all the searches within one group have been completed, in this step the candidates for the next group are determined and exceptions are processed.

In the above flowchart, steps (803) and (804) are characteristic of the authentication process of this embodiment.

[0064]

According to the present invention, when an arbitrary keyword is uttered and registered, voice characteristics (individual difference / personality / attribute) are extracted, classified according to the characteristics, and registered in the database. Next, when performing the authentication by voice inputting the same keyword as at the time of registration, the voice feature (individual difference / personality / attribute) is extracted and classified again to verify which classification frame of the database is suitable. Similar data is searched only from the optimum classification group to obtain the authentication result. The advantages of performing these processes are the advantages specific to voice authentication, such as faster data entry, less occupancy of eyes and hands, increased workability, and less skill required for data entry. In addition, since the data in a huge database is narrowed down and searched, the number of similar data is reduced as compared with the whole search, so that the reliability is increased and the search speed is increased.

[Brief description of drawings]

FIG. 1 is a conceptual diagram of a registration process according to a first embodiment.

FIG. 2 is a conceptual diagram of an authentication process according to the first embodiment.

FIG. 3 is a conceptual diagram of a registration process according to the second embodiment.

FIG. 4 is a conceptual diagram of an authentication process according to the second embodiment.

FIG. 5 is a flowchart illustrating a registration process according to the first embodiment.

FIG. 6 is a flowchart illustrating an authentication process according to the first embodiment.

FIG. 7 is a flowchart illustrating a registration process according to the second embodiment.

FIG. 8 is a flowchart illustrating an authentication process according to the second embodiment.

FIG. 9 is a block diagram illustrating the system configuration of the present invention.

[Explanation of symbols]

101 Block diagram showing raw voice of arbitrary keyword 102 Conceptual diagram of voice input device for registration such as microphone 103 Block diagram showing registration computer section 104 Data flow 105 Block diagram showing voice modeled voice data 106 Processing Step 107 of flow characteristic extraction processing 108 Direction of flow of processing flow 108 Step 109 of processing flow classification processing Step 110 of processing flow registration processing Voice database 111 Group 201 classified by characteristics Raw voice of the same keyword as at the time of registration Block diagram shown 202 Conceptual diagram of voice input device for authentication such as microphone 203 Block diagram showing computer part for authentication 204 Data flow 205 Block diagram showing voice modeled voice data 206 Step 207 of feature extraction process of process flow Processing flow Flow direction 208 Step 209 of processing flow classification processing 210 of processing flow registration processing 210 Voice database 211 Group 301 classified by feature Block diagram showing raw voice of arbitrary keyword 302 Voice input device for registration such as microphone Conceptual diagram 303 Block diagram showing computer unit for registration 304 Flow of data 305 Block diagram showing voice data modeled as voice 306 Step 307 of feature extraction processing of processing flow Direction of flow 308 of processing flow Processing classification of processing flow Step 309 of processing flow Registration step 310 Voice database 311 Group 312 classified by characteristics Conceptual diagram of work schedule for a month with a hospital 313 Conceptual diagram when classified by each working hours 401 B Registration of nurse Kiwa same as time Block diagram showing a raw voice of a voice 402 Conceptual diagram of a voice input device for authentication such as a microphone 403 Block diagram showing an authentication computer section 404 Data flow 405 Block diagram showing voice data modeled as voice 406 Features of processing flow Step 407 of extraction processing Direction of flow 408 of processing flow Step 409 of processing flow classification processing Step 410 of processing flow registration processing 410 Voice database 411 Group 412 classified by characteristics Current duty zone 413 Classified group 1001 CPU 1002 Voice input unit 1003 Storage device 1004 RAM 1005 Input unit 1006 Output unit 1007 Raw voice data 1008 Voice model

Claims (2)

[Claims] 1. In a registration-type speaker authentication system, a voice characteristic (individual difference) depending on a frequency range of a voice, a length of a keyword to be input, a contained sound element, etc. at the time of voice registration.
To extract data, to classify by voice characteristics, to store data in the classified groups, to extract voice characteristics during voice authentication, and from which classification frame to register data based on voice characteristics. To determine whether or not to search, a method to search for an approximate value from the determined classification group, a method to output the authentication result, a method to confirm the authentication result to the subject, and a method to perform a re-search based on the correctness of the authentication result. A voice registration / authentication system, comprising: 2. In a registration-type speaker authentication system, when registration time or authentication time can be managed by a schedule, a means for classifying a voice of an input person according to a schedule at the time of voice registration, and storing data in the classified group. Means, a means for classifying voices according to the time at the time of voice authentication, a means for deciding from which classification frame the registered data is to be searched, a means for searching an approximate value from the decided classification group, and a means for issuing an authentication result. And a means for confirming the authentication result with the subject and a means for re-searching based on whether the authentication result is correct or incorrect, the voice registration authentication system.